Report on accuracy evaluation of the ML model

Description of Work and Role of Specific Beneficiaries:Task 1.1: Creation of Data Dictionary and Combined machine learning approaches for task and skill-based modelling of occupationLead partner: ITCL; Timeline: M1 – M16; Total number of Person Months allocated to secondments: 20.To create a synergy between the partners and the work packages a detailed data dictionary will be composed as the first study of Task1.1. Then, some classifiers (such as Linear Discriminant Analysis, Support Vector Machine, Decision Tree, Random Forest) will be implemented with several feature selection methodologies (such as ReliefF, ILFS, HDMR, Laplacian etc.) to determine a highly accurate model for the classification of the skills and tasks either they are routine or non-routine. KHAS will be involved to the machine learning methodology development through the secondments of its ER and ESRs. ITCL will execute the pre and postprocessing of the OECD and O*SET data sets through the secondments of its technical staff. In order to assist KHAS and ITCL, ARC, ICBE and JONL will analyse the existing data sets in terms of real human information depending on their Human Resources (HR) departments.Task 1.2: Building a comprehensive deep learning model with automation probabilityLead partner: KHAS; Timeline: M1 – M20; Total number of Person Months allocated to secondments: 16.In this Task1.2, in accordance with the activities of Task1.1, deep-learning methodologies will be combined with the previous model in order to predict the tasks within the occupation models, which might disappear, continue or transform in near future. ITCL will apply different deep learning methodologies with the previous model developed in Task1.1 in order to determine the automation probability of occupations. KHAS will transform and improve the automation probability model of occupations in order to predict the occupation models, which might disappear, continue or transform in near future.Task 1.3: Testing the model Lead partner: ARC; Timeline: M10 – M36; Total number of Person Months allocated to secondments: 16.Pilot data from real employees of ARC, ICBE and JONL will be collected with a short survey which is in OECD dataset format. The results of survey will be pre-processed and used for the verification of the classifier. ARC and ITCL will develop and execute a survey in order to collect pilot data for the verification of the classifier developed in Task1.1 applied into the two different contexts (i.e. two different sectors and two different countries/cultures). ITCL will execute the pre-processing of the collected data sets through the secondments of its technical staff. KHAS will use the pre-processed survey for the verification of the classifier through the secondments of its ER and ESRs.